Optimisation de la Maintenance Conditionnelle Des Systèmes Mécaniques par Analyse Vibratoire

DJABALLAH, Said (2023) Optimisation de la Maintenance Conditionnelle Des Systèmes Mécaniques par Analyse Vibratoire. Doctoral thesis, Faculté des Sciences et de la technologie.

Text thése DJABALLAH-Final.pdf Download (11MB)

Abstract

Industrial installations are becoming increasingly complex. Monitoring their condition is essential for ensuring system safety, achieving cost savings, and enhancing productivity. This necessitates the utilization of sophisticated and highly effective monitoring techniques. In this context, the first objective of this work is to experimentally understand the phenomenon of vibrations within rotating machines, as well as to concretize of certain common defects such as imbalances and bearing faults. However, the most fundamental aspect of this work lies in the development of a new and highly effective diagnostic approach for the early and accurate detection of bearing faults, based on transfer learning. To achieve this goal, firstly, a test bench was designed and built, capable of simulating several mechanical failures. On the other hand, we developed a simple and economic data acquisition system for our project using an Arduino UNO microcontroller and an accelerometer (ADXL-345). Second, our study examines the partial knowledge transfer, for the diagnosis of bearing faults, by freezing layers in varying proportions to leverage both freeze and fine-tuning strategies. To evaluate the proposed strategy, three pre-trained models are used, namely ResNet-50, GoogLeNet and SqueezeNet. Each network is trained using three different optimizers: SGD, Adam and RMSprop. We evaluate the performance of the suggested technique in terms of defect classification rate, specificity, precision, and training time. The classification results obtained using the CWRU datasets show that the proposed technique reduces the training time while improving diagnostic accuracy, thereby improving the performance of bearing fault diagnosis

Actions (login required)

View Item